Practical sample size calculations for surveillance and diagnostic investigations.

The likelihood that a study will yield statistically significant results depends on the chosen sample size. Surveillance and diagnostic situations that require sample size calculations include certification of disease freedom, estimation of diagnostic accuracy, comparison of diagnostic accuracy, and determining equivalency of test accuracy. Reasons for inadequately sized studies that do not achieve statistical significance include failure to perform sample size calculations, selecting sample size based on convenience, insufficient funding for the study, and inefficient utilization of available funding. Sample sizes are directly dependent on the assumptions used for their calculation. Investigators must first specify the likely values of the parameters that they wish to estimate as their best guess prior to study initiation. They further need to define the desired precision of the estimate and allowable error levels. Type I (alpha) and type II (beta) errors are the errors associated with rejection of the null hypothesis when it is true and the nonrejection of the null hypothesis when it is false (a specific alternative hypothesis is true), respectively. Calculated sample sizes should be increased by the number of animals that are expected to be lost over the course of the study. Free software routines are available to calculate the necessary sample sizes for many surveillance and diagnostic situations. The objectives of the present article are to briefly discuss the statistical theory behind sample size calculations and provide practical tools and instruction for their calculation.